next up previous
Next: Typical Applications Up: Advantages of Direct Search Previous: Advantage 6: Exploring Limited


Given the potential DS advantages listed above (most of them related to partial observability), it may seem that the more ambitious the goals of some RL researcher, the more he/she will get drawn towards methods for DS in spaces of fairly general algorithms, as opposed to the more limited DPRL-based approaches.

Standard DS does suffer from major disadvantages, though, as I will point out later for the case of realistic, stochastic worlds.

Juergen Schmidhuber 2003-02-19

Back to Reinforcement Learning and POMDP page